Goldman's Solomon says AI financing demand stays strong

Goldman Sachs CEO David Solomon said demand for AI infrastructure financing remains strong despite a recent selloff in AI stocks, CNBC reported.

Solomon told CNBC’s Leslie Picker on Monday that Goldman has been “very active” in AI capital markets deals over recent weeks and expects that to continue. “There’s still very strong demand for investments in the AI infrastructure build and the technology around the growth of AI,” Solomon said.

The comments come as AI-related stocks, including chipmakers and neocloud providers, have tumbled in recent weeks. The selloff began before the midterm elections and intensified after Democrats won sweeping victories, raising concerns that political opposition could slow the industry’s expansion. Investors are reassessing whether the hundreds of billions of dollars pouring into AI infrastructure will generate returns.

Solomon acknowledged the political shift but dismissed it as a temporary recalibration. “Changes in the political environment are among factors that will create recalibrations and ups and downs, and we’re seeing some of that,” he said. “But I still deeply believe in the long-term trajectory of the power of AI in our economy to drive productivity.”

The timing is significant for Goldman. The bank is among the lead underwriters for Anthropic’s upcoming IPO, which could be one of the largest public offerings of the year. Solomon declined to comment in detail about the deal but said the market environment is “still pretty constructive.” The IPO will be a direct test of whether public-market investors share Goldman’s confidence in AI companies or the selloff has damaged the appetite.

Solomon also pushed back against political resistance to data center construction, which has emerged as a flashpoint in states where local communities oppose the power and water demands of large compute facilities. “We certainly don’t want this infrastructure going to places outside the U.S., which is what will naturally happen because the demand for compute is there,” Solomon said. The argument is that restricting data center construction domestically pushes the compute capacity overseas, where U.S. regulators have no oversight and the economic benefits do not accrue to American workers or companies.

The financing demand Solomon describes is the debt and equity capital that AI companies need to build data centers, buy GPUs, and train models. Goldman earns fees structuring that financing, whether through bond issuance for neocloud operators, equity raises for model labs, or IPO underwriting. If the demand is strong, Goldman’s revenue from AI-related deals grows regardless of what happens to the stock prices of the companies it is financing. The bank is selling the picks and shovels, not mining the gold.

The tension is between Solomon’s view from the deal pipeline and the view from the trading floor. AI stocks have fallen because investors are pricing in lower expected returns on AI infrastructure. If those investors are right, the companies building that infrastructure will need less financing over time, not more. Solomon is betting that the current selloff is a sentiment-driven overshoot, not a signal that the underlying demand for compute has weakened. The Anthropic IPO will be the first major data point on whether he is right.

What I think

Solomon is talking his book, and that is fine. Goldman is the lead underwriter on the Anthropic IPO. If the CEO of the lead bank says the market environment is constructive, he is either telling the truth or trying to make it true by saying it. Both outcomes are bullish for the IPO in the short term. The question is whether the demand he sees in the financing pipeline reflects real economic demand for compute or a financing bubble that has not yet popped.

The evidence supports Solomon on the demand side. AI model labs are reporting record usage. Enterprise adoption is accelerating. The compute requirements for training frontier models are doubling every six to nine months. None of that is consistent with a demand collapse. The selloff is about pricing, not demand. Investors bought AI stocks at valuations that assumed infinite growth and are now marking them down to reflect finite growth. The infrastructure still needs to be built, and it still needs to be financed, and Goldman is still the bank that does that.

The political risk is the one Solomon is most worried about, and it is the one he spent the most time pushing back against. Data centers are unpopular with local communities because they consume enormous amounts of power and water and create few permanent jobs. Democrats who won midterms on a populist message are not going to override local opposition to help AI companies build compute capacity. If the permitting process slows, the financing demand Solomon sees today translates into delayed projects tomorrow, and Goldman’s pipeline shrinks.

The Anthropic IPO is the test. If it prices at the top of the range and trades up, Solomon is right and the selloff was a buying opportunity. If it prices below range or breaks in the aftermarket, the financing demand he sees today is the last wave, not the next one. Goldman has every incentive to make the IPO work, and the bank has the distribution muscle to get it done. But the market sets the price, and the market has been selling AI stocks for three weeks. The IPO will tell us whether the market believes Solomon or not.

The argument that data centers should be built in the U.S. rather than overseas is correct but incomplete. The compute will be built somewhere. The question is whether the U.S. can build it fast enough to maintain the lead that American AI companies currently hold, and that question is about permitting and power grids, not about financing. Goldman can provide the capital. It cannot provide the substation permits. If the political opposition to data centers is real, the bottleneck shifts from capital to concrete, and no amount of financing demand solves a zoning problem.